Re: [PATCH] docs: submitting-patches: ask bug fixes for their user-visible impact
From: Willy Tarreau
Date: Mon Nov 16 2026 - 15:01:55 EST
On Mon, Nov 16, 2026 at 11:58:59AM -0800, Andrew Morton wrote:
> On Mon, 16 Nov 2026 14:52:05 -0500 Sasha Levin <sashal@xxxxxxxxxx> wrote:
>
> >
> > Willy pointed out the other side of this: ask an LLM for the worst case
> > and it'll stack up every config option, module and USB emulator needed
> > to get there, while for backports you also want to know whether
> > innocent users are affected at all.
> >
> > I think both fit if we ask about the two cases separately. Something
> > like this on top?
> >
> > diff --git a/Documentation/process/submitting-patches.rst b/Documentation/process/submitting-patches.rst
> > --- a/Documentation/process/submitting-patches.rst
> > +++ b/Documentation/process/submitting-patches.rst
> > @@ -63,14 +63,26 @@ from upstream, so include anything that could help route your change
> > downstream: provoking circumstances, excerpts from dmesg, crash
> > descriptions, performance regressions, latency spikes, lockups, etc.
> >
> > -For a bug fix, this means saying what the bug does to a running system:
> > -the worst that a user can see, and what it takes to trigger it. An
> > -account of what the code does wrong is not a substitute, as it leaves
> > -every reader to work the consequences out again. If the bug was found
> > -by review or by a tool and has never been seen to happen, say so. If
> > -it has no effect that a user could notice, say that too. Maintainers
> > -and the people looking after stable trees decide from this whether the
> > -fix needs to be backported.
> > +For a bug fix, this means saying what the bug does to a running system.
> > +An account of what the code does wrong is not a substitute, as it leaves
> > +every reader to work the consequences out again. Cover two cases:
> > +
> > +- Users who are not looking for the bug. Say what a normal workload is
> > + likely to run into, and separately what the worst case is and what it
> > + takes to get there: configuration, hardware, privileges, timing. If
> > + the bug was found by review or by a tool and has never been seen to
> > + happen, say so. If it has no effect that a user could notice, say
> > + that too.
> > +
> > +- Someone triggering the bug on purpose. Say whether an unprivileged
> > + user could do it and what that would take. A race that no normal
> > + workload loses can still be won by a program that retries it on every
> > + CPU for an hour. If that makes the bug a vulnerability under
> > + Documentation/process/threat-model.rst, follow
> > + Documentation/process/security-bugs.rst.
> > +
> > +Maintainers and the people looking after stable trees decide from this
> > +whether the fix needs to be backported.
> >
> > Quantify optimizations and trade-offs. If you claim improvements in
> > performance, memory consumption, stack footprint, or binary size,
> >
> > If that works for both of you I'll fold it into a v2.
>
> Looks good, thanks. Still
> Acked-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
>
> Of course we don't have to get this perfect. Close enough is good
> enough. Deploy it, observe effects, fine-tune later.
Agreed, looks great to me as well.
Reviewed-by: Willy Tarreau <w@xxxxxx>
Willy